INVESTIGATION OF THE POSSIBILITY OF FOREST TYPE MAPPING IN ARID 
AND SEMI-ARID REGIONS USING LANDSAT ETM+ DATA 
Farzin Naseri 
Assistant professor, Department of Environmental Sciences, International Center of Science & High Technology & 
Environmental Sciences, P.O. Box: 76315-117, Kerman, Iran, Email: naseri@icst.ac.ir 
Ali A. Darvishsefat 
Associate professor, Faculty of Natural Resources, University of Tehran, 
P. O. Box: 31585-4314, Karaj, Iran, Email: adarvish@chamran.ut.ac.ir 
KEY WORDS: Forest, Classification, Accuracy, Landsat, Fusion, Quality, Soil, Vegetation. 
ABSTRACT 
Landsat ETM+ data from the National Park of Khabr in Kerman province, dating May 2000, were analyzed to investigate the 
possibility of forest type mapping in arid and semi-arid regions.Quality evaluation of the image showed any radiometric distortion. 
Orthorectification was implemented to reduce relief displacement. The RMS error obtained was less than half a pixel. The ground 
truth map allocating 50% of the total area was prepared by fieldwork, using strip sampling. Different forest types considering the 
density, were qualitatively estimated in the strips based on typology definitions,. The original and synthetic bands were obtained 
applying tasseled cap transformation, PCA, ratioing and bands fusion. Furthermore, the parameters of the soil line relation were 
applied to produce suitable vegetation indices to reduce soil reflect 
ance. The best band set, based on the divergence between classes 
signatures, using sample areas were selected. Forest type classification utilizing ML, MD, PPD and SAM classifiers were performed 
to separate pure and dominant types of Pistacia atlantica, Acer monspessulanum, Amygdalus elaeagnifolia, A. scoparia and a mixed 
type. Because of spectral similarity between the pure and dominant types, these classes merged together and the classification was 
repeated. In this case the highest overall accuracy and kappa coefficient equal to 47% and 23% respectively, were achieved by MD 
classifier. Accuracy assessment and signature separability criterions showed undesirable separation between the whole forest types, 
except for Amygdalus scoparia. By merging all of the types but A. scoparia and performing the classification again, the highest 
overall accuracy and kappa coefficient equal to 92% and 68% respectively were resulted utilizing MD classifier. Based on the results, 
in such regions, low forest canopy, increases the role of background reflection and this makes undesirable results. Therefore high 
resolution sensors data and improved classification methods are advised. 
1. INTRODUCTION 
The forests in arid and semi-arid regions play an effective role 
in soil conservation and preventing destructive floods. These 
forests are ecologically important due to holding valuable 
genetic reserves and causing a rich biodiversity. Sothat it is 
necessary to image them based on comprehensive 
investigations. The starting point at this field is to achieve 
suitable and fresh information to recognize them. In this regard 
Satellite data with their own characteristics such as being able to 
cover large areas, their revisit frequency and their possibility of 
automatic analysis (Darvishsefat, 2002), can be considered to 
meet this aim. On the other hand, it seems obvious that these 
kinds of data have a great potential to diminish the fieldwork 
and cosequently, lowering the costs of forest data acquisition. 
These data are applied in different forest researches including 
typology studies (May er al., 1997; Mickelson et al., 1998; 
Huang ef al., 2001). In arid and semi-arid regions also, various 
investigations have been recently carried out using satellite data 
(Satterwhite & Henly, 1987; Leprieur et al., 1996; Hurcom & 
Harrison, 1998; Todd & Hoffer, 1998). In this study the 
potential of ETM+ data for forest type mapping in the national 
park of Khabr as a part of arid and semi-arid regions was 
investigated. 
2. STUDY AREA 
The study area measuring 1818 hectares is a part of the national 
park of Khabr in Kerman province in the south east of Iran. this 
area extends from 28? 46' 00" to 28° 49' 30" N and from 56? 28' 
30" to 56? 37' 30"E. The elevation ranges from 2000 to 
2600 meters above sea level. The main forest species are 
Pistacia atlantica, Acer monspessulanum, Juniperus 
excelsa, and Amygdalus spp. Accompnied by secondary 
species like Celtis caucasica, Cotoneaster persicus and 
Crataegus microphylla. 
3. DATA 
In this investigation ETM+ data, including 7 spectral bands 
dated 19 may 2000, were applied. Furthermore, 1:25000 digital 
topographic maps were used to determine GCPs and also to 
evaluate the precision of geomatric correction. 
4. Methods 
4.1 Ground truth map 
In this research, based on the present possibilities, a strip ground 
truth map, allocating 5096 of the total area measuring 909 ha 
was prepared. Forest types were estimated qualitatively in the 
strips, based on typologic definitions (Gorgi Bahri, 2000). Large 
numbers of recognized forest types and small area of some of 
them, made the types to be generalized and consiquently 9 
forest types including 4 pure types (Pistacia atlantica, Acer 
monspessulanum, Amygdalus | Spp., and Am. Scoparia) 4 
dominant types (Pistacia atlantica, Acer monspessulanum, 
Amygdalus spp., and Am. Scoparia) and a mixed type were 
concluded. In these types if the canopy area of a species was 
more than 90% of the total, a pure type and if it was between 
50% to 90% a dominant type was resulted. If the canopy area of 
the first species was less than 50% of the total and the second 
410 
Int 
spe 
Obi 
On 
gre 
usi 
In 
cla 
cel 
the 
  
  
4.2 
[t i 
situ: 
(Dai 
radit 
dupl 
Were 
Orth 
digit 
(Ant 
size 
fores 
creat 
pring 
band 
In a 
MSA 
prod 
(Ter 
The 
Figure